Simplify configurable optimizer API - #518
Conversation
facebook-github-bot
added
the
CLA Signed
jrapin
left a comment
jrapin
left a comment
There was a problem hiding this comment.
Would something like this be more natural to use?
| budget: Optional[int] = None, | ||
| num_workers: int = 1, | ||
| scale: float = 1.0, | ||
| diagonal: bool = False |
There was a problem hiding this comment.
Specify extra configuration parameters with defaults here (as you usually do)
|
|
||
| class ParametrizedCMA(base.ParametrizedFamily): | ||
| """TODO | ||
| class ParametrizedCMA(base.ConfiguredOptimizer): |
There was a problem hiding this comment.
Then just create a Wrapper class
| MilliCMA = ParametrizedCMA(scale=1e-3).with_name("MilliCMA", register=True) | ||
| MicroCMA = ParametrizedCMA(scale=1e-6).with_name("MicroCMA", register=True) | ||
| CMA = ParametrizedCMA().set_name("CMA", register=True) | ||
| DiagonalCMA = ParametrizedCMA(diagonal=True).set_name("DiagonalCMA", register=True) |
There was a problem hiding this comment.
And decline the different values of the extra parameters
jrapin
left a comment
jrapin
left a comment
There was a problem hiding this comment.
I also updated an example you are familiar with: SplitOptimizer3,5,9,13 :D
| super().__init__(parametrization, budget=budget, num_workers=num_workers) | ||
| if num_vars: | ||
| if num_optims: | ||
| if num_vars is not None: |
There was a problem hiding this comment.
@teytaud make if num_vars can lead to unexpected behavior if what you wanted is to test agains None, so always use if num_vars is not None
| num_vars: Optional[List[Any]] = None | ||
| ) -> None: | ||
| super().__init__(parametrization, budget=budget, num_workers=num_workers, num_optims=num_optims, num_vars=num_vars) | ||
| SplitOptimizer3 = ConfSplitOptimizer(num_optims=3).set_name("SplitOptimizer3", register=True) |
There was a problem hiding this comment.
This is all it takes now to have all the variants
|
Agreed by @teytaud , merging and following with other PRs to propagage |
2 participants
Types of changes
Motivation and Context / Related issue
The simplified process means:
How Has This Been Tested (if it applies)
Checklist